Postmortem: A Modern Approach to Managing Kubernetes Clusters
summary
Conto replaced Cluster Autoscaler with Carpenter to reduce node fragmentation and reclaim capacity, runs Argo CD GitOps using an app-of-apps bootstrap to auto-provision stacks across six new EKS clusters, and templates Kubernetes manifests with Elm to shrink configs from about 100 lines to a dozen. The result manages ~20,000 pods across 15+ clusters with ~200 daily deployments, cut maintenance from one week to three days, reduced incidents from five to two per month, and achieved a 55% VM activity cost reduction; Carpenter and Argo CD are the core technologies.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 2:09Complexity and cost
- 7:44Cluster scaling with Carpenter
- 10:50Simplifying configs
- 13:16GitOps with Argo CD
- 18:27GitOps bootstrapping
- 22:21Automatic tool deployment
- 27:57Command Ops and upgrades
- 31:03GitOps results
- 33:08Multicluster resilience
- 36:41Multicloud interfaces
- 37:43Improvement review
- 41:37Q&A
related talks
32mDon't Start With 500 Clusters: Patterns in Scaling Multi-Cluster Kubernetes…
36mOptimizing Resource Usage in Kubernetes
36mMulti-Cluster Kubernetes Quick Delivery with IaC and GitOps
39mOptimizing Resource Usage in Kubernetes
22mHarnessing Karpenter: Transforming Kubernetes Clusters with Argo Workflows
31mSaving Money Exposed Our Architecture - Escaping ClusterAPI with Karpenter
22mKarpenter Efficient scaling of Kubernetes clusters
29mRunning Kubernetes on 'small scale' clusters - lessons learned